This is a nice thought experiment. We all know it proves nothing and at the same time, formally, it does prove something. Let's get back to PG's original essay

This is a nice thought experiment. We all know it proves nothing and at the same time, formally, it does prove something. Let's get back to PG's original essay. What he says essentially is, if the goal of high-level languages is to make computer instructions (programs) shorter, then making them even shorter might be the ultimate goal of any high-level language. The shorter the better. Sounds like a very valid point. However, there's something I think PG omits or disregards - the complexity of the language in terms of the total number of abstractions introduced, which intuitively should be low as well. In Lisp, which itself is a succinct language, you may have a framework with some number of functions in one case, and an equivalent framework with fewer functions in another case. Same applies to built-in constructs and notions - you may have fewer notions that cover same scenarios or vice versa, both producing (roughly) the same length in terms of tokens. But clearly, there's a difference and we'd prefer the framework that has fewer functions/abstractions/notions to remember. So if we agree on this, Arc' is bad because it introduces more notions while achieving no or very little effect, if not harm actually.